EA isn't a superorganism where everyone (and all organizations) share the same empirical assumptions (or normative views). This complicates the analysis because money held by people who do grantmaking one doesn't consider important should already be discounted.

edit: In the current funding landscape, my point isn't very relevant because it seems relatively easy to get funding if someone's project looks strong on some plausible assumptions. However, in the future, we could imagine scenarios where large sums of money are deployed for ambitious strategies where some people think the strategy is good and others think it might be too risky. Example: Buying compute for an AI company where some funders think the company has enough of a safety mindset, while other funders would prefer to wait and evaluate. 

At what level of payoff is that bet worth it? Lets say the bet is a 50/50 triple-or-nothing bet. So, either EA ends up with half its money, or ends up with double. I'd guess (based on not much) that right now losing 50% of EA's money is more negative than doubling EA's money is positive.

There is an actual correct answer, at least the abstract. According to the Kelly criterion, on a 50/50 triple-or-nothing bet, you should put down 25% of your bankroll.

Say EA is now at around 50/50 Crypto/non-Crypto, what kind of returns would justify that allocation? At 50/50 odds, there's actually no multiple that makes the math work out.

But that's just for the strict case we're discussing. See the section on "Investment formula" for what to do about partial losses.

Finally, instead of a 50/50 triple-or-nothing bet, we can model this as a 75/25 double-or-nothing bet (same EV as you bet). In that case, we get that a 50/50 allocation is optimal.

But note that the Kelly criterion is optimizing for log(wealth)! Log(wealth) approximates utility in individuals, but not in aggregate. Since EA is trying to give all its money away, the marginal returns slope off much more gradually. (See some very rough estimates here.) If you're just optimizing for wealth, you would be okay with a riskier allocation.

BTW, it's not just "over-invested in X", you have to think about the entire portfolio. So given that almost all EA money is either Sam or Dustin, you have to consider the correlation between Crypto and FB stock.

I'll also add that you have to consider all future EA money in determining what % of the bankroll we're using.

It doesn't really matter though, since EA doesn't "own" or "control" Sam's wealth in any meaningful way.

I think you should have made this post a question. It being a post made me think you actually had an answer, so I read it, and was disappointed you didn’t actually conclude anything.

Are you talking about individuals who are EA, or funds?

Mainly funds, but also individuals (especially wealthy individuals)

Imagine we were coordinating among everyone doing patient philanthropy (patient philanthropy being any situation where you keep wealth instead of liquidating it and giving it away to the current best opportunity that can use the money) and then optimising the overall portfolio. We're asking the question: In what ways would we choose a different portfolio? I'm just explaining one of the considerations which is the diminishing marginal utility of the whole pool.

[...] bonus points if they can get (legal or informal) commitments from the donors. And then maybe use this information to hedge against individual donors who they believe are making irrational trades, or else are too far out on the risk curve. (Then again should funds devoted to EA have the authority to decide this top-down, or is delegating these choices to donors better?)

I agree, this is exactly what I'm looking for from these comments! This is a pretty reasonable way that we can coordinate on the overall portfolio.

For anyone reading, here's how this would work. Let's say Fund X thinks Rich Individual Y is overly bullish on crypto, and that Fund X's inside view is that EA should only be 20% invested in crypto. Fund X, instead of investing 20% of their funds in crypto, might want to put all their money in other investments instead, to bring the overall EA portfolio closer to 20%.

In general, if people reveal their donation intentions and investment portfolios, we can do a better job hedging against each other to optimise the overall portfolio.

Also, if we come up with a diminishing marginal returns curve, people can judge the optimality of a given portfolio better. @AppliedDivinityStudies has posted a link and thoughts on this part of the discussion.

"[...] afaik EA Funds doesn't invest their money at all"

Wow - This seems pretty bad, unless in practice all their holding are temporary (ie. They go down to approximately $0 more than once a year). I'm pretty that the optimal portfolio doesn't include much cash, given that cash appreciates in real value at the wonderful interest rate of negative 2-3% (negative 6% last year?)